Abstract
A fundamental challenge for the visual system is perception of the social world. A key component of social perception is the ability to categorize human interactions based on bodily form and motion, in a manner that is independent of the particular actors involved, the viewpoint, or setting. Here we use fMRI to identify brain regions that represent human interaction categories. We scanned participants with fMRI while they viewed short video clips of two-person interactions (slap, kick, shove, bite, pull, brush, massage, tap). For each action, several factors were crossed: the actors, their roles (Agent, Patient), the direction of action (left-to-right, right-to-left), and the scene (four indoor backgrounds). We focused on several regions of interest that respond to features relevant for distinguishing actions, such as bodies (EBA and FBA), motion (hMT+), and biological motion (pSTS). We also examined activity in early visual cortex (EVC) and the parahippocampal place area (PPA) as control regions. Multivariate pattern analysis revealed that the action category (e.g. kick) could be decoded from activation patterns in EVC, EBA, FBA, hMT+, and pSTS, but not PPA. Representations of actions in these regions were invariant to changes in actors, roles, and scenes, and invariant to action direction in all regions except EVC. Furthermore, a dissociation between action information (EBA, FBA, hMT+, pSTS) and scene information (PPA) was found, consistent with functional specialization of these regions. In contrast, EVC contained information that distinguished the stimuli on many visual dimensions (action, action direction, and scene). Our results reveal that the neural patterns in brain regions previously shown to respond to bodily form and motion contain information that differentiates categories of actions, invariant to visual elements such as the actors, action direction, and setting.
Meeting abstract presented at VSS 2015